Copyright startuppedia
"Pixa AI, a Jaipur-based artificial intelligence startup, has launched Luna, a speech-to-speech foundational voice AI model that doesn"t just talk, but can sing, whisper, pause, and do emotion-driven interaction in real time.Advertisment Unlike traditional systems that convert speech to text and back, Luna AI can reportedly process audio and generate human-like speech, removing conversion latency and enabling faster, more expressive, emotionally intelligent conversations. It can whisper, pause, sing, and respond in context, making a level of human-like emotional nuance and responsiveness. About Pixa AI Founded by 25-year-old IIT-BHU alumnus Sparsh Agrawal, the AI startup positions Luna as India"s first voice model that directly converts speech into speech, bypassing text layers used by current systems. The Jaipur firm has received early backing from Kunal Shah (Cred), Shankar Narayana (ex-MD, Carlyle Asia Growth Partners), and Kunal Kapoor (celebrity and investor) and Nikhil Kamath"s WTFund, where Agrawal was part of the accelerator programme last year. All about Luna “When we were building voice AI toys for kids, we realised that an emotional-first AI is missing from the market...something expressive, something you actually want to talk to,” said Agrawal. “That"s what led us to create Luna...a model that doesn"t just speak, but can sing, talk to you, and is the first AI you"d want to talk back to.” Currently, Luna can converse in English and even comprehends tonal and dialectal variations across different geographies. Pixa AI plans to introduce multilingual support within the next two to three months, covering 12 major Indian languages and additional global ones, bringing the total to over 30. The Jaipur startup"s core team currently comprises four full-time members, with plans to expand to about 10 in the coming months, primarily on the technology side. The firm intends to actively participate in the IndiaAI Mission. According to the founder, they are discussing GPUs, model development, and data sets with government officials. To scale Luna"s multilingual capabilities, Pixa is in talks with the government to secure GPU access under the mission. “We believe the future of voice AI is not automating call support centres but building artificial emotional intelligence,” said Agrawal. “Humans relate to sarcasm, tone, and emotion. Luna focuses on expressivity, multiple accents, tonal variety, entertainment-first design, and not monotonous robotic voice output.” By eliminating the text conversion layer, Pixa claims Luna achieves 600-millisecond end-to-end latency, compared to 900–1200 milliseconds in typical speech-to-text-to-speech systems. The voice AI model also detects tone shifts, such as anger or excitement, and adjusts its delivery accordingly. “It"s a pure voice-to-voice model...speech in, speech out,” Agrawal explained. “It detects tonality in real-time if you"re angry or calm, and adjusts its tone, pace and delivery dynamically.” The company is currently running proof-of-concept pilots with an Indian automobile manufacturer and a US-based wellness startup, where Luna AI powers interactive storytelling and in-car entertainment. The brand is also intending to enter the customer-service sector, seeking emotionally adaptive sales or support voice agents. Also read: Lenskart IPO opens today for subscription: Check GMP, price band, and key details (startuppedia.in) "
